The productivity paradox: evidence from indirect indicators of service sector productivity growth

Transcription

1 The produtivity prdox: evidene from indiret inditors of servie setor produtivity growth E D W A R D N. W O L F F New York University 1. Introdution Diffiulties in mesuring the output of servie setors hve een well doumented in mny studies. This hs led to orresponding diffiulties in reting relile index of produtivity in servie industries. On the other hnd, input mesures re quite dequte in servie setors, s in other industries within the eonomy. Lor, pitl, nd mteril inputs re esily identifile nd mesurle in servies, nd re, in priniple, no different thn in other industries. The si prolem, then, is how to mesure produtivity in n industry in whih output is diffiult to mesure ut inputs re reltively esily mesured. 1 Most reent ttempts to otin etter mesure of produtivity in servie industries hve imed t improving the output mesures. For exmple, in the nking industry, the numer of heks proessed or lered per hour hs een proposed s n output inditor; for the irline industry, pssenger-miles; for legl servies, the numer of wills prepred or the numer of rel estte losings per yer; nd for the helth industry, the numer of proedures performed per yer (see, for exmple, Bresnhn, Milgrom, nd Pul 1992; Den nd Kunze 1992; Fixler nd Zieshng 1992; nd Gordon 1992). In lmost ll ses, the use of these diret indies of servie output results in higher mesured produtivity growth thn those sed on onventionl ntionl ounting dt. There re prolems in these pprohes due to the ft tht most servies produe omposite output. Suh pprohes usully pture only one or severl spets of the output nd often the lest importnt prts of the industry s 2 tivity. For exmple, in the nking industry, the most importnt tivity in terms of revenue nd personnel is the lon deprtment. It is very diffiult to rrive t suitle index for the output of lon deprtment, though there re spets of the tivities of lon deprtment tht n e redily monitored. These inlude the numer of lon pplitions reeived, the numer of lons pproved, the dollr mount of lons pproved, the interest rte hrged, other fees generted, the defult rte, nd so on. In lw firms, the hief revenue soure re the omplex legl ses tken on nd the most suitle mesure is, perhps, hours illed -- vlue dded mesure. Cndin Journl of Eonomis Revue ndienne d Eonomique, Vol. 32, No. 2 April / vril Printed in Cnd Imprimé u Cnd / 99 / $1.50 Cndin Eonomis Assoition

2 282 Edwrd N. Wolff An lterntive pproh used in this study is to onsider severl indiret inditors of produtivity growth in the servie setor y exmining hnges in the input mix. The trik is to void using servie output mesures or prie defltors in designing suh indies. It should e noted t the outset tht the indiret mesures relte to the growth rte of produtivity rther thn to its level. Moreover, the indies developed provide irumstntil evidene rther thn diret evidene on produtivity movements. Two pprohes re used here. The first is sed on hnges in diret inputoutput oeffiients. I utilize hnges in the inter-industry oeffiients nd the pitl-lour nd mterils-lour rtios s indies of produtivity growth. Tehnologil progress (totl ftor produtivity or TFP growth) in goodsproduing industries hs historilly een ssoited with not only rising level of output per unit of input ut lso with hnge in input proportions -- in prtiulr, with inresing pitl-lour nd mterils-lour rtios. Goodsproduing setors with low produtivity growth, on the other hnd, hve typilly een hrterized y reltively stle rtio of pitl to lour s well s mterils used per lour hour. The seond pproh onsiders hnges in the ouptionl omposition of employment within servie setors. The justifition is similr. A numer of indies of ouptionl hnge re used here: (i) the hnge in the overll ouptionl omposition of employment within n industry; (ii) the rtio of sientifi nd tehnil personnel to totl industry employment; (iii) the proportion of knowledge workers in totl employment; (iv) the proportion of professionls nd tehnil lour to totl employment; nd (v) the proportion of mngeril nd dministrtive lour to totl employment. A relted inditor is hnges in the verge ognitive skill level of the workfore in n industry. Skill levels n e mesured y the verge level of edutionl ttinment or y the Ditionry of Ouptionl Title (DOT) diret skill mesures. For the nlysis of oeffiient hnges, I rely on U.S. input-output dt for the yers 1958, 1967, 1977, nd The nlysis of hnges in ouptionl omposition will e sed on dt from the U.S. deennil Censuses of Popultion for 1960, 1970, 1980, nd The reminder of the pper is orgnized s follows. The next setion (setion 2) develops the ounting frmework. Setion 3 desries the dt soures nd methods. Setion 4 presents desriptive sttistis nd setion 5 the regression results. Conluding remrks re mde in the lst setion. 2. Aounting frmework Define: 3 U = n input or use ommodity-y-industry flow mtrix, where u ij shows the totl mount of ommodity i onsumed y industry j,

3 Produtivity prdox 283 V = n output or mke industry-y-ommodity flow mtrix, where v ij shows the totl output of ommodity j produed y industry i, 4 T x = V 1 = (olumn) vetor, showing the gross output of eh ommodity, where supersript T refers to the trnspose of the mtrix, nd 1 = vetor with unit entries. Also, let: T y = (V - U)1 = (olumn) vetor of finl demnd y ommodity, e = (row) vetor, showing totl employment y industry, nd k = (row) vetor, showing totl pitl stok y industry. To derive the orresponding tehnil oeffiients, I will mke use of the ommodity tehnology model, where it is ssumed tht eh ommodity is produed y the sme tehnology, irrespetive of the industry of prodution. In this se, industries re onsidered independent omintions of outputs j, eh with their seprte input oeffiients ( ). The ommodity tehnology ij requirements (oeffiient) mtrix is given y: T -1 A = U[V ] = mtrix of interindustry tehnil oeffiients, (1) where V is restrited to squre mtrix (tht is, there re s mny industries s 5 ommodities). Row vetors of lour nd pitl stok oeffiients n e derived in nlogous fshion: T -1 R = e[v ] = (row) vetor of lour oeffiients, nd T -1 ê = k[v ] = (row) vetor of pitl oeffiients. In ddition, let us define n = totl employment ( slr) in the eonomy, = totl pitl stok ( slr) in the eonomy, w = the nnul wge rte ( slr), ssumed onstnt ross industries, r = the rte of profit on the pitl stok ( slr), lso ssumed onstnt 6 ross industries, nd p = (row) vetor of ommodity pries, given y the Leontief eqution: -1 p = (wr + rê)(i - A). In the I-O frmework, setorl output is mesured y gross ommodity output x (lterntely lled gross domesti output or GDO), while the inputs onsist of employment, fixed pitl, nd mterils (intermedite inputs). The rte of TFP (totl ftor produtivity) growth for setor j is given y: ð j / -(Óp i id ij + wdr j + rdê j)/p j (2) 7 where ð is the orresponding row vetor nd d refers to the differentil. Sine produtivity growth rtes re mesured over disrete time periods rther thn instntneously, the verge vlue shres of p i, w, nd r over the smple period re used to mesure ð. 8 Aggregte TFP growth, ñ, is defined s ñ / [pdy - wdn - rd]/py. (3) This mesure is diretly nlogous to eqution (2), the index of setorl TFP,

4 284 Edwrd N. Wolff exept tht intermedite inputs re netted out. 3. Dt soures nd methods Our si dt soure onsists of U.S. input-output dollr flow tles, whih were originlly otined from the Bureu of Eonomi Anlysis on the 87-setor level for yers 1958 nd 1967 in single-tle formt, nd on the 85-setor level 9 for yers 1967, 1977, nd 1987 in the dul use-mke tle formt. The 1967, , nd 1987 dt re ville in seprte mke nd use tles. There were severl djustments required to mke the four tles omptile, whih re desried in detil in Wolff (1997). 11 All mtries re deflted to 1972 dollrs using setorl prie defltors. Produtivity growth rtes for re lulted using the single-tle si frmework (nd mking use of the 1967 single tle dt). Produtivity growth rtes for nd re lulted using the use-mke frmework (nd relying on the 1967 dul tle dt). Beuse of lignment diffiulties etween the vrious input-output yers (severl industries re ollpsed in the 1987 tle, in prtiulr), produtivity growth estimtes re ville for only 68 industries. The 85 industries re divided into two groups: goods nd servies. The goods 12 industries inlude griulture (1-4), mining (5-10), onstrution (11-12), mnufturing (13-64), trnsporttion (65), ommunitions (66-67), nd utilities (68). Servies inlude trde (69), finne, insurne, nd rel estte (70-71), government servies (78, 79, nd 82), nd ll other servies (72-77 nd 84). It should e stressed tht I inlude ommunitions, trnsporttion, nd utilities in the goods setor euse for the purposes here they hve reltively esily mesured output nd re more like the other goods industries thn servies from this stndpoint. Employment dt for 267 ouptions nd 64 industries re otined from the deennil Census of Popultion for yers 1960, 1970, 1980, nd Sine ouption nd industry lssifitions hve hnged sustntilly with eh ensus, I used Commere Deprtment omptiility tles for nd to produe onsistent mtries for the four yers. Fortuntely, there were only very minor hnges in lssifition etween 1980 nd 1990 (see Wolff 1996 for more detils on the onstrution of these mtries). The mesure of ognitive skill is sed on the fourth (1977) edition of the Ditionry of Ouptionl Title (DOT). For some 12,000 jo titles, it provides vriety of lterntive mesures of jo-skill requirements sed upon dt 13 olleted etween 1966 nd I use s n index of ognitive skill the mesure Sustntive Complexity (SC), whih is omposite (ftor nlyti) mesure of skills developed y Roos nd Treimn (Miller et l. 1980, Appendix F), refleting the edutionl nd trining time requirements of the jo, s well s the needs for synthesizing, oordinting, nd nlyzing dt nd the generl lerning, resoning, verl, nd numeril ility of the jo. This mesure is

5 Produtivity prdox 285 developed for eh of the 267 ouptions (see Wolff 1996 for more detils). Another mesure of ognitive skills, whih is derived from the 1970 Census of Popultion dt, is Medin Yers of Shooling-1970 (EDUC-1970). Medin yers of shooling is omputed for eh ouption in 1970 on the sis of tul shooling ttinment reported y respondents in the 1970 Census of Popultion. If the tul skill requirements of eh ouption remin onstnt over time, then EDUC-1970 serves s n inditor of the hnges in the edutionl requirements of the workple. Averge industry skill sores re omputed s weighted verge of the skill sores of eh ouption, with the ouptionl employment mix of the industry s weights. Computtions re performed for 1960, 1970, 1980, nd 1990 on the sis of the ouption y industry employment mtries. Another dimension of ouptionl skills is sed on the numer of knowledge produers in n industry. The si dt re gin from the U.S. Deennil Censuses. In the lssifition shem, professionl nd tehnil workers hve generlly een lssified s knowledge workers, depending on whether they re produers or users of knowledge. The line is somewht ritrry t points, nd judgment lls hve een mde (see Wolff 1996 for more detils). 4. Desriptive sttistis 4.1 Trends in produtivity growth Before presenting the regression nlysis, it is helpful to see wht the onventionlly mesured produtivity growth figures re for the vrious setors of the eonomy. As shown in pnel 1 of tle 1, the nnul rte of lour produtivity growth for the entire U.S. eonomy fell from 1.8 perent over to 0.9 perent over nd then to 0.7 perent over In the goods industries, there ws generlly slowdown in lour produtivity growth etween the nd periods nd then modest reovery in the period. This ws true for ll setors exept nondurle mnufturing, whose produtivity growth rte remined strong throughout the three periods nd utilities, where produtivity growth ws exeptionlly high in the first two periods nd then fell to virtully zero in the third. Altogether, lour produtivity growth in the goods-produing industries (inluding ommunitions, trnsporttion, nd utilities) verged 2.5 perent per yer in the period, fell to 1.7 perent per yer in the period, nd then reovered slightly to 1.9 perent per yer in the period. The pttern is very different for the servie industries. Lour produtivity in wholesle nd retil trde ws strong in the period (2.0 perent per yer), turned negtive in the next period, nd then reounded in (1.5 perent per yer). In finne, insurne, nd rel estte (FIRE, for short) nd in generl servies, lour produtivity growth dropped etween nd nd then turned negtive in the period. For the government setor,

6 286 Edwrd N. Wolff TABLE 1 Lour nd totl ftor produtivity (TFP) growth y mjor setor, (verge nnul growth in perentge points) Lour Produtivity Growth A. Goods Industries Agriulture, forestry, nd fisheries Mining Constrution Mnufturing, Durles Mnufturing, Nondurles Trnsporttion Communition Eletri, gs, nd snitry servies B. Servie Industries Wholesle nd retil trde Finne, insurne, nd rel estte Generl Servies Government & government enterprises Aggregted Setors Totl Goods Totl Servies Totl Eonomy (GDP) Totl Ftor Produtivity Growth A. Goods Industries Agriulture, forestry, nd fisheries Mining Constrution Mnufturing, Durles Mnufturing, Nondurles Trnsporttion Communition Eletri, gs, nd snitry servies B. Servie Industries Wholesle nd retil trde Finne, insurne, nd rel estte Generl Servies Government & government enterprises Aggregted Setors Totl Goods Totl Servies Totl Eonomy (GDP) See eqution (2) for the definition of setorl TFP growth nd eqution (3) for the definition of overll TFP growth.

7 Produtivity prdox 287 lour produtivity growth remined virtully zero etween 1958 nd Overll, lour produtivity growth in servies fell from 1.2 perent per yer in to 0.46 perent per yer in nd to 0.35 perent per yer in As result, the gp in nnul lour produtivity growth etween the goods setors nd the servie setors widened, from 1.25 perentge points in to 1.50 perentge points in Pnel 2 shows the orresponding results for TFP growth. TFP growth for the totl eonomy ws very strong in the period, verging 1.5 perent per yer, nd then fell shrply to 0.3 perent per yer in nd, unlike lour produtivity growth, showed no reovery in the period. In the goods setor s whole, TFP growth verged 2.1 perent per yer in , fell shrply to 0.6 perent per yer in , nd then reounded to 1.0 perent per yer in In ontrst, in servies, nnul TFP growth fell from 0.9 perent in the first period to 0.2 perent in the lst two periods. Time trends in TFP growth for the individul setors re, in generl, quite similr to the overll pttern for their group. All goods-produing setors experiened slowdown in TFP growth etween the nd periods nd ll exept mining, nondurle mnufturing, ommunitions, nd utilities show reovery in the period. TFP growth in ll four servie setors turned from positive to negtive etween nd nd, with the exeption of trde, delined even further in By oth mesures, produtivity growth ppers to e muh lower in servies thn in the goods-produing industries. Moreover, the disprity hs generlly widened over time, etween 1958 nd Indeed, in the period, oth lour produtivity nd TFP growth in the servie industries were quite respetle, verging out one perent per nnum. However, y the period, produtivity growth in servies ws virtully zero. Severl eonomists hve ontended tht the pprent poor performne of servies in more reent yers is due to inresing prolems in the mesurement of their output over time; not due to tul hnges in produtivity. I now onstrut some relted mesures of tehnologil tivity to investigte this issue. 4.2 Mesures of tehnologil tivity Severl mesures of tehnologil tivity re developed tht do not diretly rely on setorl output mesures. The first is the growth in the rtio of pitl to lour, shown in the top pnel of tle 2. If lour produtivity growth is, in relity, higher in goods industries thn servies, prt of this might e ounted for y lrger inrese in the pitl-lour rtio. Some support is provided for this rgument. Over the entire period, the pitl-lour rtio grew y 2.4 perent per yer in the goods industries nd 1.5 perent per yer in servies. However, the pitl-lour rtio tully inresed fster in servies thn in goods industries in the period nd t out the sme rte in the period. Moreover, pitl-lour growth ws higher in trde nd in FIRE thn in

8 288 Edwrd N. Wolff TABLE 2 Growth rtes of the pitl-lour nd interindustry inputs-lour rtios y mjor setor, (verge nnul growth in perentge points) Cpitl / Lour Rtio A. Goods Industries Agriulture, forestry, nd fisheries Mining Constrution Mnufturing, Durles Mnufturing, Nondurles Trnsporttion Communition Eletri, gs, nd snitry servies B. Servie Industries Wholesle nd retil trde Finne, insurne, nd rel estte Generl Servies Government & government enterprises Aggregted Setors Totl Goods Totl Servies Totl Eonomy Interindustry Inputs / Lour Rtio A. Goods Industries Agriulture, forestry, nd fisheries Mining Constrution Mnufturing, Durles Mnufturing, Nondurles Trnsporttion Communition Eletri, gs, nd snitry servies B. Servie Industries Wholesle nd retil trde Finne, insurne, nd rel estte Generl Servies Government & government enterprises Aggregted Setors Totl Goods Totl Servies Cpitl is sed on gross pitl stok, lour on employment, nd interindustry inputs re in onstnt (1972) dollrs.

9 Produtivity prdox 289 mnufturing. A similr rgument pplies to the rtio of totl intermedite inputs to lour. 14 Over the entire period, the rte of growth in this rtio ws greter in the goods thn in the servie setor, ut the differene ws not gret (2.3 versus 1.8 perent per yer). Moreover, this rtio grew muh fster in servies thn in the goods setor in the period. Another nomly is tht the rte of inrese in the rtio of totl intermedite inputs to lour over ws out the sme in trde nd generl servies s in mnufturing. Tle 3 shows two inditors of investment tivity. The first of these is investment in offie, omputing, nd ounting equipment (OCA) per full-time equivlent employee (FTEE). In the period--the period when purhses of OCA were y fr the gretest--fire led the wy t $1,068 (in 1987 dollrs) per FTEE, followed y mining ($523), utilities ($464), durles mnufturing ($266), nd ommunitions ($226). As whole, the servie setor invested more intensively in omputer equipment thn the goods setor, ut this ws lrgely due to the investments mde y FIRE. Trde nd generl servies were tully elow verge in terms of OCA investment per FTEE. The seond inditor is totl investment in equipment, mhinery, nd instruments (inluding OCA) per FTEE. It should e noted t the outset tht totl equipment investment ws more thn ten times greter thn OCA investment, even in the period, whih proly explins why omputeriztion y itself hs not hd muh effet on overll produtivity growth. The goods industries invested muh more hevily thn the servie setor in equipment per FTEE -- out doule overll. The leding setors were ll goods produers -- utilities, ommunitions, nd mining. Another inditor of the rte of tehnologil tivity is the degree to whih the interindustry oeffiient struture shifts over time. For this, I employ n index of similrity. First define: t t t s j = ij / Ó i ij, (4) whih shows the input (in onstnt dollrs) from industry i to industry j s shre of the totl sum of interindustry inputs (ll in onstnt dollrs) into setor j. Then, the stndrd similrity index for industry j for time periods t1 nd t2 is given y: t1 t2 Ó i s j@s j 12 SI = ))))))))))))))))). (5) t1 2 t2 2 1/2 [Ó i (s j Ó i (s j ) ] t1 t2 The index SI is the osine etween the two vetors s nd s nd vries from 0 -- when the two vetors re orthogonl, to 1, when the two vetors re identil. 15 The index of dissimilrity, DI, is defined s: DI = 1 - SI. (6)

10 290 Edwrd N. Wolff TABLE 3 Averge nnul investment in offie, omputing, nd ounting equipment (OCA) nd in mhinery, equipment, nd instruments per FTEE, (1987$) Investment in OCA / FTEE A. Goods Industries Agriulture, forestry, nd fisheries Mining Constrution Mnufturing, Durles Mnufturing, Nondurles Trnsporttion Communition Eletri, gs, nd snitry servies B. Servie Industries Wholesle nd retil trde Finne, insurne, nd rel estte , Generl Servies Government & government enterprises NA NA NA NA Aggregted Setors Totl Goods Totl Servies (exept government) Totl Eonomy (exept government) Investment in Mhinery & Equipment / FTEE A. Goods Industries Agriulture, forestry, nd fisheries 3,816 6,875 5,134 5,143 Mining 7,416 11,769 14,401 11,600 Constrution 2,290 2,540 1,504 2,027 Mnufturing, Durles 1,841 2,580 3,089 2,540 Mnufturing, Nondurles 2,258 3,291 3,825 3,145 Trnsporttion 4,931 7,058 6,117 6,013 Communition 9,140 14,314 18,253 14,429 Eletri, gs, nd snitry servies 11,658 18,970 21,309 17,872 B. Servie Industries Wholesle nd retil trde 957 1,349 1,934 1,512 Finne, insurne, nd rel estte 2,796 4,691 8,766 6,202 Generl Servies 1,682 2,056 1,991 1,928 Government & government enterprises NA NA NA NA Aggregted Setors Totl Goods 3,060 4,509 4,880 4,191 Totl Servies (exept government) 1,433 2,054 2,834 2,292 Totl Eonomy (exept government) 1,995 3,340 3,732 2,613 Soure for Investment Dt: Bureu Of Eonomi Anlysis, Deprtment of Commere, Diskette of Detiled Investment y Industry. Soure for FTEE: Bureu of Eonomi Anlysis, Ntionl Inome nd Produt Aounts, diskettes.

11 Produtivity prdox 291 TABLE 4 Dissimilrity indies (DI) of interindustry tehnil oeffiients nd the distriution of ouptionl employment, Dissimilrity Index (DI) for Interindustry Coeffiients A. Goods Industries Agriulture, forestry, nd fisheries Mining Constrution Mnufturing, Durles Mnufturing, Nondurles Trnsporttion Communition Eletri, gs, nd snitry servies B. Servie Industries Wholesle nd retil trde Finne, insurne, nd rel estte Generl Servies Government & government enterprises Totl Goods Totl Servies Dissimilrity Index (DI) for Employment A. Goods Industries Agriulture, forestry, nd fisheries Mining Constrution Mnufturing, Durles Mnufturing, Nondurles Trnsporttion Communition Eletri, gs, nd snitry servies B. Servie Industries Wholesle nd retil trde Finne, insurne, nd rel estte Generl Servies Government & government enterprises Totl Goods Totl Servies All Industries Interindustry inputs re in onstnt (1972) dollrs.

12 292 Edwrd N. Wolff A greter vlue of the index DI indites more dissimilrity etween the two vetors. Results for DI re shown in tle 4. The ommunitions setor ws y fr the most dynmi in terms of shifting its input struture over the period of 1958 to 1987 ( DI vlue of 0.54), prtiulrly for the period The seond most dynmi setor ws generl servies (0.26), followed y durles mnufturing (0.20), utilities (0.18), nd trnsporttion (0.15). However, overll, goods industries experiened more hnge in their interindustry oeffiient struture thn servies, prtiulrly in the period. Pnel 2 of tle 4 presents mesures of the dissimilrity index DI sed on hnges in ouptionl struture over eh dede. It should first e noted tht the degree of overll ouptionl hnge ws greter in the 1980s (DI equls 0.10) thn in the 1960s (0.06) nd muh greter in these two periods thn in the 1970s (0.02). These results onfirm nedotl evidene out the sustntil degree of industril restruturing during the 1980s. Moreover, though the degree of ouptionl hnge ws greter in the goods thn the servie setor, the differene ws reltively smll (0.20 versus 0.16 over the period ). The three setors tht experiened the gretest ouptionl restruturing over the three dedes were ommunitions (0.26), utilities (0.23), nd generl servies (0.23). Ouptionl hnge ws prtiulrly low in griulture (0.03), onstrution (0.08), nd trnsporttion (0.09). I use two other tehnologil vriles. The first of these is RDSALES j, the mount of R&D expenditure in onstnt dollrs per onstnt dollr of net sles in setor j. The dt re otined from the Ntionl Siene Foundtion (vrious yers). These dt hve historilly een ville only for the mnufturing 16 setor. As shown in tle 5, RDSALES in mnufturing hs remined reltively onstnt over time, t lest in omprison to the wide flututions in lour produtivity nd TFP growth. Indeed, the weighted verge of RDSALES for overll mnufturing shows slight dip etween the nd periods ut lmost no hnge etween the ltter period nd The rtio of R&D to sles ws onsiderly higher in durle mnufturing thn nondurles (lmost ftor of three) nd in the period rnged from low of 0.4 perent in food produts to high of 18.3 perent in irrft. The seond is n index of diret tehnologil spillovers from the supplying to the purhsing setor, given y: o TFPIND jt / Ó i ijt@ð it, whih is mesure of setor j's indiret knowledge gin from tehnologil o hnge in its supplying setors. The mtrix A is identil to the A mtrix exept tht the digonl is set to zero to prevent doule-ounting of TFP growth. It is ssumed tht the informtion gined from supplier i's TFP is proportionl to its importne in setor j's input struture. In Wolff (1997), this vrile ws found to e highly signifint in explining industry TFP growth.

13 Produtivity prdox 293 TABLE 5 The rtio of reserh nd development expenditures to sles (RDSALES) for mnufturing industries, (figures re in perent) Mnufturing, Nondurles Food produts Too produts Textiles, pprel, fris Lumer, wood produts, furniture Pper nd pper produts Printing nd pulishing Chemils nd llied produts Petroleum refining, relted produts Ruer nd misellneous plstis Footwer, lether, lether produts Glss, stone nd ly produts Misellneous mnufturing Mnufturing, Durles Primry iron nd steel Primry nonferrous metls Frited metl produts Engines, turines, nd mhinery Computer nd offie equipment Servie industry mhinery Applines; eletri equipment Motor vehiles Airrft nd prts Other trnsporttion equipment Sientifi instruments Ophthlmi & photogrphi equipment Ordnne nd essories Unweighted Averge Mnufturing, Nondurles Mnufturing, Durles Mnufturing, Totl Weighted Averge Mnufturing, Nondurles Mnufturing, Durles Mnufturing, Totl Weighted y gross output (GDO).

14 294 Edwrd N. Wolff As shown in tle 6, TFPIND for the eonomy s whole followed the sme trend s overll TFP growth, flling shrply etween the nd periods nd then reovering slightly in the lst. This is not surprising sine TFPIND is weighted verge of TFP growth rtes of individul industries. However, wht is surprising is tht wheres TFPIND for the goods setor s whole lso onformed to this pttern, TFPIND for servies fell etween oth the first nd seond nd the seond nd third periods s well (turning negtive in the ltter). This is due to the ft tht servie industries tend to uy from eh other nd tht goods industries lso tend to uy from eh other. Also, s result of this, TFPIND ws generlly higher in goods industries thn servie industries. TABLE 6 Knowledge spillovers etween setors s mesured y the rte of TFP growth of supplying setors (TFPIND), (figures re in perent) A. Goods Industries Agriulture, forestry, nd fisheries Mining Constrution Mnufturing, Durles Mnufturing, Nondurles Trnsporttion Communition Eletri, gs, nd snitry servies B. Servie Industries Wholesle nd retil trde Finne, insurne, nd rel estte Generl Servies Government & government enterprises Totl Goods Totl Servies All Industries Figures re unweighted verges of individul industries within eh mjor setor. 3.3 Mesures of skill level Tles 7 nd 8 show vrious indies of rin-power y industry. The first of these is the rtio of knowledge workers to totl industry employment (tle 7). The servie industries s group were more intensive in their use of knowledge workers thn the goods setor ut the leding setor ws ommunitions (21.8 perent in the 1980s), followed y the government setor (16.1 perent), generl servies (15.6 perent), nd FIRE (15.4 perent). The inrese in the shre of knowledge workers in totl employment etween 1960 nd 1990 ws out the sme for servies s for the goods industries.

15 Produtivity prdox 295 TABLE 7 Knowledge workers nd relted workers s shre of industry employment y mjor setor in the U.S., (figures re in perent) Chnge Knowledge Workers / Totl Employment A. Goods Industries Agriulture, forestry, nd fisheries Mining Constrution Mnufturing, Durles Mnufturing, Nondurles Trnsporttion Communition Eletri, gs, nd snitry servies B. Servie Industries Wholesle nd retil trde Finne, insurne, nd rel estte Generl Servies Government & government enterprises Aggregted Setors Totl Goods Totl Servies All Industries Sientists, Computer Anlysts, Engineers, & Tehniins / Totl Employment Totl Goods Totl Servies All Industries Professionl nd Tehnil Workers / Totl Employment Totl Goods Totl Servies All Industries Mngers nd Administrtors / Totl Employment Totl Goods Totl Servies All Industries

16 296 Edwrd N. Wolff TABLE 8 Averge skill sores y mjor setor in the U.S., % Chnge SC (Sustntive Complexity) A. Goods Industries Agriulture, forestry, nd fisheries Mining Constrution Mnufturing, Durles Mnufturing, Nondurles Trnsporttion Communition Eletri, gs, nd snitry servies B. Servie Industries Wholesle nd retil trde Finne, insurne, nd rel estte Generl Servies Government & government enterprises Aggregted Setors Totl Goods Totl Servies All Industries Med.Edu-70 (Medin Edution-1970) A. Goods Industries Agriulture, forestry, nd fisheries Mining Constrution Mnufturing, Durles Mnufturing, Nondurles Trnsporttion Communition Eletri, gs, nd snitry servies B. Servie Industries Wholesle nd retil trde Finne, insurne, nd rel estte Generl Servies Government & government enterprises Aggregted Setors Totl Goods Totl Servies All Industries

17 Produtivity prdox 297 The rtio of sientists, omputer nlysts, engineers, nd tehniins to totl employment ws muh greter in the goods setor thn in servies (5.2 versus 2.4 perent in 1990) nd the rtio grew fster in the goods setor thn in servies over the three dedes. In ontrst, the totl numer of professionl nd tehnil workers s shre of totl employment ws more thn twie s gret in servies s in the goods industries in 1990 nd grew fster in servies thn the goods industries etween 1960 nd The shre of mngers nd dministrtors in totl employment ws greter in servies ut grew fster in the goods setor. As shown in tle 8, ognitive skill levels (SC) were, on verge, higher in the servie setor thn the goods setor. In 1990, employees in FIRE hd the highest verge SC sore (5.36), followed y generl servies (4.91), ommunitions (4.86), nd the government setor (4.68). The growth in men SC ws out the sme in servies s in the goods industries etween 1960 nd The pttern is very similr for the Med.Edu-70 (Medin Edution-1970) sore. The verge Med.Edu-70 sore ws higher in servies thn the goods setor nd ws led y generl servies (13.4 in 1990), followed y FIRE (13.2), ommunitions (13.0), nd government (12.9). The perentge hnge in this sore over the three dedes ws lso out the sme in the goods nd servie setors. 5. Regression nlysis I now turn to regression nlysis to sort out the influenes of these vrious tehnologil inditors on mesured produtivity growth mong industries. There re three min questions of interest. First, whih, if ny, of these ftors is found to hve signifint effet on mesured produtivity growth? Seond, is the goodness of fit etter mong goods-produing industries lone in omprison to ll industries, inluding servies? Third, do the regression results differ sustntilly etween goods-produing nd servie industries, nd, if so, does this differene throw ny light on mesurement prolems in servie output? The dependent vrile of the regressions is the rte of TFP growth denoted y TFPGRT. The independent vriles inlude the tehnologil inditors desried ove, suh s the rtio of OCA investment to FTEE; the rtio of totl equipment investment to FTEE; the ouptionl hnge index; the rtio of R&D expenditure in onstnt dollrs to GDP in onstnt dollrs, RDSALES; weighted sum of TFP growth rtes for supplying industries, TFPIND; sientifi nd tehnil personnel s proportion of totl industry employment, nd oth the level nd hnge in verge industry skill sores. The vriles used nd their symols re listed in tle 9. The si estimting eqution is: TFPGRT = + RDSALES + TFPIND + TECHACT + å (7) j 0 1 j 2 j 3 j j where TECHACT j is one of the other indies of tehnologil tivity in the industry nd å j is stohsti error term. It is ssumed tht the å jt re

18 298 Edwrd N. Wolff independently (ut not neessrily identilly) distriuted. The regression results reported in tles 10, 11 nd 12 use the White proedure for heterosedstiity-onsistent ovrine mtrix. The onstnt 0 is usully interpreted s the pure rte of tehnologil progress. The smple is pooled ross-setion time-series dt set onsisting of 68 industries nd 3 time periods ( , , nd ). From Grilihes (1980), the oeffiient of RDSALES is interpreted s the rte of return to R&D, under the ssumption tht the (verge) rte of return to R&D is equlized ross setors. Time dummies for the periods nd re introdued to llow for period-speifi effets on produtivity growth not ttriutle to R&D or the other tehnologil inditors. A dummy vrile identifying the 10 servie industries is inluded to prtilly ontrol for mesurement prolems in servie setor output. The regressions were lso run seprtely for the group of 58 goods-produing industries nd for the group of 10 servie industries. TABLE 9 Vrile key TFPGRT: Annul rte of growth of totl ftor produtivity. RDSALES: Rtio of R&D expenditure in onstnt dollrs to sles in onstnt dollr. TFPIND: A weighted sum of the nnul rtes of TFP growth of supplying industries, where the weights re given y the interindustry input oeffiients. OCAFTEE: Investment in OCA, in 1987 dollrs, per FTEE. DIOCCUP: Dissimilrity Index sed on 267 ouptions y dede. SERVDUM: Dummy vrile for the 10 servie industries. KNOWLAVG: The rtio of knowledge workers to totl employment (period verge). KNOWLCHG: The hnge in the rtio of knowledge workers to totl employment over the period. PROFAVG: The rtio of professionl nd tehnil workers to totl employment (period verge). PROFCHG: The hnge in the rtio of professionl nd tehnil workers to totl employment over the period. ADMINAVG: The rtio of dministrtive nd mngeril workers to totl employment (period verge). ADMINCHG: The hnge in the rtio of dministrtive nd mngeril workers to totl employment over the period. SCAVG: Averge SC (sustntive omplexity) sore y industry, period verge. SCCHG: The hnge in verge SC (sustntive omplexity) sore y industry over the period. MEDUCAVG: Averge Med.Edu-1970 sore y industry, period verge. MEDUCCHG: The hnge in verge Med.Edu-1970 sore y industry over the period.

19 Produtivity prdox 299 TABLE 10 Regressions of industry TFP growth (TFPGRT) on tehnologil vriles: ll industries Speifition Independent Vriles (1) (2) (3) (4) Constnt (1.44) (1.58) (2.17) (1.95) RDSALES (1.99) (1.97) (1.81) (1.82) TFPIND (2.77) (2.45) (2.82) (2.61) OCAFTEE (0.79) (0.38) DIOCCUP (1.71) (1.21) SERVDUM (1.23) (1.75) (1.37) (1.85) 2 R R Std Err ó Smple Size Estimted oeffiients re shown elow the respetive independent vriles nd the solute vlue of the t-sttisti is shown in prentheses. The White proedure for heterosedstiityonsistent ovrine mtrix is used in the estimtion. See tle 10 for vrile key. Signifint t the.01 level (two-tiled test). Signifint t the.05 level (two-tiled test). Signifint t the.10 level (two-tiled test). The first set of regression results, for ll industries, is shown in tle 10. In speifition (1), whih inludes only RDSALES, TFPIND, nd servie dummy vrile (SERVDUM), oth RDSALES nd TFPIND hve positive oeffiients nd re signifint t the five nd one perent level, respetively. The oeffiient of SERVDUM is negtive, s expeted, ut not signifint here. The goodness 2 2 of fit, s mesured y the djusted-r (R ), is only When dummy vriles re dded in for the two time periods (DUM6777 nd DUM7787), neither is sttistilly 2 signifint nd the R sttisti delines (results not shown). In speifitions (2)-(4), OCAFTEE (investment in OCA, in 1987 dollrs, per FTEE) nd DIOCCUP (the dissimilrity index sed on hnges in the ouptionl omposition of employment within the industry) re inluded s independent vriles long with RDSALES nd TFPIND. Both vriles hve negtive oeffiients. However, OCAFTEE is not sttistilly signifint, wheres DIOCCUP is signifint t the 10 perent level when inluded y itself

20 300 Edwrd N. Wolff ut not signifint when inluded with OCAFTEE. The two other tehnology vriles -- DIACOEFF (the dissimilrity index sed on totl interindustry oeffiients) nd EQUIPFTE (investment in equipment, mhinery, nd instruments, in 1987 dollrs per FTEE), -- hve negtive ut insignifint oeffiients. In tle 11, I dd in the vrious inditors of the degree of rinpower within industry. Four of the indies -- KNOWLAVG (the shre of knowledge workers in totl industry employment, verged over the period), PROFAVG (the shre of professionl nd tehnil workers), SCAVG (the verge sustntive omplexity or ognitive skill sore of the industry, verged over the period), nd TABLE 11 Regressions of industry TFP growth (TFPGRT) on tehnologil nd skill vriles: ll industries Speifition Independent Vriles (5) (6) (7) (8) (9) Constnt (1.71) (2.12) (0.06) (0.70) (0.26) RDSALES (1.76) (1.73) (1.83) (1.48) (1.23) TFPIND (2.76) (2.47) (3.19) (2.28) (2.39) DIOCCUP (1.68) (1.66) (2.52) (2.05) (1.92) Averge skill (0.47) (0.30) (2.20) (0.15) (0.40) Chnge in skill (0.42) (0.41) (0.52) (1.40) (1.13) SERVDUM (1.23) (1.26) (2.08) (1.22) (1.40) 2 R R Std Err ó Smple Size Skill KNOWLAVG PROFAVG ADMINAVG SCAVG MEDUCAVG Mesures KNOWLCHG PROFCHG ADMINCHG SCCHG MEDUCCHG See tle 10 for vrile key. Signifint t the.01 level (two-tiled test). Signifint t the.05 level (two-tiled test). Signifint t the.10 level (two-tiled test).

21 Produtivity prdox 301 MEDUCAVG (the verge medin edution-1970) -- re eh sttistilly insignifint (see speifitions 5 to 9). In two ses, the oeffiients re tully negtive (though insignifint). The hnges in these vriles (s well s their nnul growth rte, whih is not shown) re mostly sttistilly insignifint, though their oeffiients re generlly positive. The mjor exeption is ADMINAVG (the shre of mngeril nd dministrtive workers in totl employment), whih hs positive oeffiient tht is signifint t the five perent level. However, the hnge in the shre of mngeril nd dministrtive workers hs n insignifint oeffiient. The evidene seems to suggest tht, with the exeption of mngeril workers, the presene of high ognitive skill workers is not prtiulrly enefiil to the produtivity growth of n industry. 17 Another interesting result is tht the oeffiient of TFPIND remins highly signifint nd its oeffiient vlue remins virtully unhnged even with the inlusion of these skill vriles. On the other hnd, RDSALES eomes less signifint nd its oeffiient vlue flls somewht. The reson for this is tht there is positive orreltion etween the level of R&D expenditures of n industry nd the industry's verge skill level (orreltion oeffiients of 0.34 etween RDSALES nd KNOWLAVG, 0.37 etween RDSALES nd PROFAVG, nd 0.34 etween RDSALES nd MEDUCAVG), refleting the sientists nd engineers employed in R&D tivity. However, the orreltion etween RDSALES nd ADMINAVG is slightly negtive (-0.09). It is lso of 2 interest tht the djusted-r sttisti flls somewht with the ddition of these skill vriles, gin with the exeption of ADMINAVG nd ADMINCHG. When the smple is restrited to goods industries only (tle 12), the oeffiient estimtes nd signifine levels of the two mjor tehnologil vriles -- RDSALES nd TFPIND -- remin virtully unhnged, s shown in speifition (10). However, none of the other tehnologil vriles is signifint, inluding OCAFTEE, DIOCCUP, DIACOEFF, nd EQUIPFTE, though they ll still generlly hve negtive oeffiients (see speifitions 11 nd 12, for exmple). It is lso of interest tht when DIOCCUP is dded to the regression, RDSALES remins signifint, though t the 10 perent level (result not shown) ut when ny of the skill mesures re dded, with the exeption of ADMINAVG, RDSALES eomes insignifint. The reson is tht mong goods industries y themselves, the orreltion etween RDSALES nd the vrious skill mesures is quite strong ( vlue of 0.54 etween RDSALES nd PROFAVG nd 0.51 etween RDSALES nd MEDUCAVG, for exmple). None of the rinpower vriles re signifint, exept ADMINAVG, whih is positive nd signifint t the 10 perent level (see speifitions 11 nd 12). Also of interest, the goodness of fit of the vrious regression forms did not improve when the smple ws restrited to the goods industries. The regression results for the servie industries y themselves, using onventionlly mesured TFP growth, re very different thn those for the goods industries. As shown in speifitions of tle 12, the oeffiient of

22 302 Edwrd N. Wolff TFPIND is now negtive nd generlly signifint t the ten perent level. The oeffiients of oth OCAFTEE nd DIOCCUP re negtive nd signifint t the 18 2 one or five perent level in every se. The djusted-r sttisti is muh higher for the servie industry regressions thn the goods industry regressions -- in the rnge of 0.20 to 0.26 ompred to 0.07 to Of ll the skill vriles, only ADMINAVG hs positive nd signifint oeffiient (t the five perent level). PROFAVG is signifint t the five perent level, ut negtive. None of the other skill level vriles is signifint. TABLE 12 Regressions of industry TFP growth (TFPGRT) on tehnologil nd skill vriles: goods nd servie industries seprtely Goods Industries Only Servie Industries Only Independent Vriles (10) (11) (12) (13) (14) (15) Constnt (1.41) (0.92) (0.31) (3.97) (0.68) (1.61) RDSALES (1.99) (1.05) (1.77) TFPIND (2.64) (2.80) (2.97) (1.86) (0.19) (1.91) OCAFTEE (5.22) (2.75) (5.42) DIOCCUP (1.03) (1.28) (2.43) (2.68) (3.48) Averge skill (1.25) (1.86) (2.21) (2.15) (1.43) Chnge in skill (0.01) (0.10) (1.01) (1.16) (1.15) 2 R R Std Err ó Smple Size Skill PROFAVG ADMINAVG PROFAVG ADMINAVG MEDUCAVG Mesures PROFCHG ADMINCHG PROFCHG ADMINCHG MEDUCCHG See tle 10 for vrile key. Signifint t the.01 level (two-tiled test). Signifint t the.05 level (two-tiled test). Signifint t the.10 level (two-tiled test).

23 6. Conlusions nd interprettion of results Produtivity prdox 303 The regressions from the ll-industry smple hve provided some striking results. First, omputeriztion (OCAFTEE) does not pper to exert positive effet on produtivity growth. Indeed, its oeffiients hve generlly een negtive (though not signifint). My results differ from those reported y Brynjolfsson nd Hitt (1997), Lehr nd Lihtenerg (THIS ISSUE), nd Ger, Gu, nd Lee (THIS ISSUE), who generlly find positive nd signifint oeffiients on their omputeriztion vriles. There seem to e three resons. First, Brynjolfsson nd Hitt, nd lso Lehr nd Lihtenerg use the firm s the unit of oservtion wheres I use the industry (Ger, Gu, nd Lee lso use industry dt). Seond, the smpling frmes re different. In prtiulr, my smple ords greter weight to servie industries. In prtiulr, finne nd insurne re mrked outliers with, y fr, the highest levels of omputeriztion nd mong the lowest in terms of TFP growth -- result I hd ommented on in erlier work (see Wolff, 1991). Third, nd perhps most importntly, I use pooled time-series nd rosssetion dt, where the other three ppers rely only on firm or industry rosssetionl dt. Sine my figures show huge inrese in omputeriztion (OCAFTEE) over time, from 1958 to 1987, nd slowdown in produtivity growth over the sme period, the regression results fil to yield signifint oeffiient on OCAFTEE. However, when I restrit the smple to the period (nd lso to goods industries), I do find positive nd signifint (t lest t the 10 perent level) oeffiient on OCAFTEE. Seond, mjor restruturing of tehnology, s refleted in hnges in the ouptionl omposition of employment (DIOCCUP), seems to hve retrdnt effet on produtivity growth. The first two sets of results might reflet the high djustment osts ssoited with the introdution of new tehnology. The prdigmti shift from eletromehnil utomtion to informtion tehnologies might require mjor hnges in the orgniztionl struture of ompnies efore the new tehnology n e relized in the form of mesured produtivity gins (see, Dvid 1991, for greter elortion of this rgument). Some onfirmtion of this hypothesis is provided y Brynjolfsson nd Hitt, for exmple, who find tht omputeriztion hs positive effet on firm-level produtivity only s long s there re onomitnt hnges in firm orgniztion. Third, rinpower, s refleted in the vrious mesures of worker skill, does not pper enefiil to produtivity growth. The lone exeption is the presene of dministrtors nd mngers (though not neessrily their growth over time), whih is positively relted to industry produtivity growth. The likely rtionle for this result is tht mny professionl workers, suh s lwyers, ountnts, dvertising personnel, nd rokers re involved in tivities tht re inherently rent-seeking in nture or unprodutive nd do not inrese mesured output. For exmple, study y Murphy, Shleifer, nd Vishny (1991) reported rossountry regression results showing tht per pit GDP growth is negtively